Procedural content and emergent gameplay, a brainstorming

[Zomg]

For the devil's advocates in this thread, keep in mind that no one in this thread is arguing that this should supercede or outmode handmade content. If four companies made four Fallout-type dialogue RPGs a year, I'd probably buy those four games every year for the rest of my life. However, the hardcore procedural-emergent route is also interesting and seemingly viable, and some of the games that I've played that precurse such an experience are among my all-time favorites. Plus, I can see a lot of possibilities in an eventual procedural-emergent and handmade convergence.

 

[galsiah]

I'm sure the order of all this makes perfect sense. I just can't see it anymore .

I'm just questioning why there's so much resistance to the idea of conceptualizing RPG characters as characters, rather than simulated actors in a simulated world. Though maybe this topic should be in its own thread.

I agree if we're talking about an RPG with 40 NPCs in it. Central characters (if there are any) will need to be hand crafted as characters for the forseeable future. However, if we're talking about a game with 1000 or more NPCs, it simply isn't possible to hand craft the character of each to a high degree of quality (though this doesn't mean that some can't be hand crafted). That's where I think emergent "quests" etc. are useful.

Allanon:
First of all, using the word "fun" is only helpful if you mean what the rest of the english-speaking world means. I understand what you meant now, so that's fine for this discussion. I still think you'd do a lot better in future by using "fun" where you want to imply cheerfulness / amusment / laughter..., rather than general purpose entertainment.
I'm not trying to be offensive here - you're clearly a lot better at languages than I am. I just think that using "fun" in the way you did is misleading for most people.

Too bad you've decided to concentrate on this nuance, instead of debating other points in my post.

I debated over everything I had time to or had something to say about. I just happen to think that in a discussion on game design, definitions of "fun" or "entertainment" are important - since that is the aim of the entire process.

Actually, I didn't address all your points, to be fair:

And that's just for his wares! Just think about the overall number of answers he should be able to answer from all possible topics.

So long as you subdivide things well, there is no trouble with having the potential for a merchant to answer 100 questions. Most of these will be generic, and will apply to all NPCs, many will apply only to merchants - but to all merchants. You certainly wouldn't be writing 100 replies per NPC, you'd be writing perhaps 1000 replies covering 200 reply types, with variables introduced to cover asking about different towns / items / NPCs etc.

Trivial? No. Possible? Yes.
How long does it take to think up and write "Yes, I get %d deliveries of %s per %s."? 15 seconds? To do that 1000 times and get reasonable wording for all replies, might take a few days. You might end up writing many more than 1000 replies. It might take one person a week or two. That's hardly monumental development time to cover a dialogue system for every generic NPC in the game.
You could also give NPC specific dialogue to any NPC, of course, or even over-ride the standard responses for specific characters.

My personal opinion is that failure to complete a task can never be rewarding, especially taking into consideration that it might have been unsolvable in the first place.... I seem to believe that reward is a major part of role playing experience, and when you decide to delve into unrewarding activities just for the sake of it, it takes out the fun (entertainment).

Material reward is one incentive for an activity, but clearly not the only one. Would OB be fun if you had to walk up and down a room repeatedly, and piles of gold appeared each time you reached an end? The material reward is important, but it is trivial compared to the satisfaction / challenge / entertainment the player gets from the process of playing.

If as a player I play a ranger, chasing a master thief through the wilderness, I might use many skills along the way. Perhaps I'll examine tracks. I might use disturbances in wildlife to give me clues. Maybe I'll use powers of divination to help me. I might ask NPCs - perhaps trusting their assertions, of reading more into them than is obvious. I might even get into combat with the thief.

If after all that, the thief gets away, do I think "I just wasted an hour for no reason."? Not necessarily at all. The chase might have been exciting and challenging. I might have used cunning tactics to persue my target. My skills and ingenuity might have been taxed all the way just to stay with him. Whether or not I catch him is of little importance, so long as I enjoyed the process. I've already got my reward through the challenge, the excitement, learning as a character, learning as a player...

The idea that an emergent quest is worthless to the player if it turns out to be too hard for him is nonsense. Perhaps catching the thief above was too hard for my character, but I can certainly have had a rewarding experience along the way. Of course it should more often be possible to succeed - at least where success seems a reasonable expectation to the player. That doesn't make occasional failure a bad thing.

Frankly, from the gameplay point of view, I see no reason to introduce quests that aren't solvable, unless we are talking about specific simulation of world events.

You wouldn't be introducing unsolvable quests. You'd be creating a situation where some tasks might turn out to be impossible. Usually it would be clear to the player when a task would be difficult or impossible.

Again, there is no real difference between setting yourself the target of "Killing creature X before it gets away", or the task of "Helping NPC X by achieving Y". Perhaps the creature will be too hard to kill, and perhaps the task will be too hard to complete. You just can't be successful at everything you want to do all the time: you are not superman, deal with it.

Your argument might have some force if RPG designers created "arbitrary" triggers, but that hasn't been my experience. If you have examples, feel free to share.

The triggers always choose the solution (or set of solutions) the designer thought of. That decision is arbitrary in the sense that there are almost always alternatives that might occur to some players, but are not included in the design. It will be impossible for a designer to script for every possibility, so some solutions get excluded.

Examples:
In Neverwinter Nights there was some quest involving a "werewolf hunter", who was actually a werewolf. To expose him, you had to do X, Y and Z until your character realised he was a werewolf, then to confront him.
I guessed that he was the werewolf before I had much evidence - though I had found some. The game did not let me confront the guy until I'd got the remaining evidence, since not all the triggers had been met.

I don't think I'm the only person to have suspected or been fairly sure that the guy was the werewolf before my character knew - it's a very obvious possiblity, particularly when the guy in question is so smug. The fact that I knew, but my character didn't, disconnected me from my character. This is a bad thing.

In Morrowind, an early quest involved finding Fargoth's hiding place. I presumed that the advice (i.e. watch where he hides it from the lighhouse) was just one useful way to proceed. I searched in various places, and watched where he went for a while, and a few days passed while I did various other things.
I definitely searched in that tree stump in the swamp because I thought it looked suspicious. I didn't necessarily think that was Fargoth's hiding place - just that something useful might be there. Did I find anything? No.

Fargoth only hides his stuff when you trigger this by watching at night from the top of the lighthouse. This is a totally arbitrary trigger, in the sense that it makes no sense in the game world. True, you are told to watch him from there, but so what? Why does it make any sense to think he'll only put anything in his hiding place when you watch him from that particular position? Even if you watch him 24 hours a day with a constant effect invisibility enchantment, he'll never hide anything there - until you watch from the 100% arbitrary position you've been told to.

Also, it is quite reasonable, even if I'd thought everything through carefully, that my character wouldn't follow the instructions he'd been given.
If you were told by someone you'd just met and had no reason to trust:
"Go to this particular isolated spot at night where you won't be seen by anyone, and wait there."
Is it not reasonable to think it might be a trap? Perhaps the guy giving you the quest just enjoys mugging unsuspecting newcomers at the top of a lighthouse, then pushing them to their deaths.


There are many more examples of this sort of thing. Pretty much any quest with a few triggers puts artificial constraints (beyond the rules of the game world) on the way the quest can be completed. If you can't think of such examples, I don't think you're trying very hard.
The important difference with emergent "quests" is that there are no constraints beyond game world constraints. If you can't achieve something, it is always because the world is like that - not bacause a designer didn't think you'd do things in the way you did.

Yes, I was thinking if it's even practical as well. I think it was galsiah who told that it is. However, the level of interactivity between npcs should be so high, we might need an extension cards for AI, just like PhysX does for physics.

Why? There being a lot of possible interactions doesn't make a difference if most NPCs don't use 95% of them most of the time - and they wouldn't. Complex physics needs to be evaluated every frame, and it is difficult to fudge calculations convincingly. Complex AI might need re-evaluation for most NPCs perhaps every 30 seconds (in the absense of important external triggers). It is also quite possible to approximate things, to skip calculations if things become too complex, etc.

Good AI is definitely a headache for the programmer, but there is no necessity for new hardware, since approximations and simplifications are possible where they wouldn't be for graphics or physics. It's certainly not easy to get right, but most of the difficulty would be in design, not in processing power.

I believe generating "satisfy general NPC need X" quest out of NPC needs would generally be single step quest.

Less often if the NPC has always tried and failed first (or been too scared... to try). Usually, if an NPC can't manage to do something himself, there will be a significant obstacle. If the obstacle is trivial, then the "quest" may often be over very quickly, e.g.:
NPC: I need money for a new spade, but can't offer anything in return.
Player: Here you go. / Sorry, I don't have any on me. / Sod off.
NPC: Thanks, I won't forget this. / Ok then. / No need to be rude.

Clearly it'd be preferable if <NPC has no money> is not usually the only obstacle. Part of "trying to achieve X himself" could involve working and saving money, if X is not immediately urgent.

99% of them would not be very compelling in terms of quest structure.

That's rather a bold statement, before anything has been tried. In any case, hero's don't usually feel compelled to go around solving mundane problems for every NPC they pass in the street. The player can easily refuse to help if a task seems boring. This will usually be obvious too - if an emergent "quest" looks very simple and uninteresting, it probably is. If it looks complex, it might be.

For instance, "bring me some X from town Y" is unlikely to be that compelling, whereas, "Prevent army X from reaching town Y" might be a bit more demanding and complex (this might be to satisfy a town / faction need, rather than an NPC need). [though as I've said before, very large scale interactions would need to be carefully controlled and balanced].

...this is necessary so, because in order to generate more intricate quest structure the interactions between NPCs and the world would have to be simulated at an increasingly exponential detail level, which is computationally unattractive and effort vs return diminishes very quickly.

Why? What does "increasingly exponential" mean in any case? Why does "effort vs return diminish"? Have you tested this?

How is a wide variety of possible interactions necessarily "computationally unattractive"? That's only true if there is no efficient way to choose between the options. Usually there will be many ways to cut out many classes of options without much computation at all. Also, most NPCs won't need to re-evaluate their situation very often - unless exceptional circumstances occur (such as attacks etc.).

In any case, there is nothing to suggest that intricate "quests" require very detailed NPC interactions. A combination of several relatively simple possibilities can quickly create an intricate situation. Take chess, or even better, Go. The individual interactions are very simple to model, but the situations created are complex.

The place you'd run into computational difficulties would not be in the simulation itself, but in getting the NPCs to solve versatile problems. This needn't necessarily be an obstacle. In most circumstances, NPC problems will be simple to solve. Where they are not, if calculations become too complex, the NPC can simply give up, or take a possibly helpful action and hope for the best. I'm sure there would be complications with this, but it might be workable.

the more you allow NPC and world interactions, the more it can breakdown. imagine after 20 hours of gameplay player arrive at most cities deserted: NPCs killed off each other, wandered off to forest to search for food, etc.
to prevent this, you ended up simulating more world rules, which in turn breaks down in more unexpected manner... and you coded more rules to prevent that, and...and so on. I think you get what i mean.

The fact that this can happen does not mean that it will. Introducing an extra rule has the potential to complicate things in new ways, but this does not always happen. Most extra rules complicate things, but happily intelligent designers do not add new rules at random. A thoughtfully introduced rule can often simplify or control a situation.

The idea that this sort of simulation will often lead to NPC killing sprees doesn't convince me. Most NPCs should be very reluctant to get into combat with anyone, unless they are attacked first. Presuming that most NPCs can find food and water even when something gets in the way of their normal strategy, killing sprees should be rare.

Bethesda examples of NPCs attacking and killing others because e.g. the other had the first NPC's rake, just seem to imply that the settings are not good. Killing should be a last resort (for most NPCs), and should only usually be an option if the NPC would otherwise die. This is a good example where a few extra rules can simplify matters - so long as the NPC has other ways to get food, water, and enough gold to live, without his rake, there is no need to kill for it. Ideally, of course, there would be a way to ask for it back, or to barter for it...

No matter how sophisticated the algorithms behind content generation systems they remain mere labour saving devices and nothing more. Such systems’ output replaces superior human authored content allowing the world builders to increase scale due to lowered resource costs. As an end user I shouldn't care how the content is generated but rather what kind of content - in terms of both quality and quantity - is delivered.

I don't think it makes sense to call an emergent system based on NPC needs a "mere labour saving device".

It is not possible to hand craft the amount of quests (i.e. an infinite amount) that are produced by generation on the fly. Therefore it makes no sense to say that you have "speeded up" the process of quest generation, because you end up in a totally different situation. If you ended up with 10000 quests, you'd have speeded things up. When you end up with an infinite amount, you have created something different.

Also, with the example of "quests" arrising out of NPC needs, the individual quests are different too. There are no trigger systems, or artificial constraints beyond the rules of the game world. Everything just emerges from the way the game world functions - for good and bad. It is not possible to script quests like this, since covering every possibility is not possible. In this case, authored content is not "superior", since it will miss possibilities that an emergent "quest" will automatically allow.

The way you describe things would be true for an automated quest generation tool that generated quests during development. It does not apply when quests are generated on the fly (since the fact that you get an infinite amount makes a real difference), and certainly not when the quests are emergent, since this created a very different quest structure.

Emergent quests are not just generated in a different way - the results as the player experiences them are very different (or rather would be very different).

 

[cmagoun]

What about semi-emergent gameplay?

(Note: I really don't want to play devil's advocate in, or crap on, this thread. It might seem like it at first glance. Please bear with me.)

A discussion a friend and I had about the ideal CRPG led to similar thoughts. What if we could track the state/needs/relationships between the NPCs in our virtual world and have them act on those needs? Quests, dialog and events would be determined by the needs and acts of NPCs.

We putzed around in general terms for a couple of weeks and finally, my friend introduced me to an acquaintance of his from the local university who was interested in computer modeling of societies and human interactions. All three of us had lunch and discussed at length these same ideas for several hours.

Then it hit me... we were talking about a massive programming problem that, even if we solved it, would probably only produce the most rudimentary of storylines. Worse, even if we solved it well, and could create a believable virtual world (we would have won some kind of Nobel Prize, but I was not thinking in those terms), the very nature of it would insure that the player would miss most of it. Sure King Albert went to war with King Bertram, but you were too far away to take part and only heard a rumor.

So, after talking to this very smart professor, I realized, we weren't looking for the same thing. He wanted to make a simulated world and I wanted... well I wasn't sure at the time. So I thought about it and realized that I wanted something that simulated a tabletop RPG campaign more than I wanted a simulated world.

I wanted the computer to GM for me.

So I thought about what I do when I GM. I don't keep a detailed tab on all of the NPCs in the game, decide on their needs and desires, and then adjudicate their actions to create events in which the players can participate. Usually, I take the characters' current situation, construct a scenario around that situation, and use the various NPCs and other elements of my setting to support that scenario. In my mind, emergent gameplay works from the environment to a situation to the player, while most tabletop RPGs work from the players to a situation to the environment (if that makes any sense).

Wheee.... I just described Morrowind!!! Good for me.

Seriously, there are tons of games that have NPC actions, dialog, and events scripted for the players to see and interact with. By dumping the idea of AI interactions causing events, I have just reverted to what RPGs have right now. However, these don't quite fit the bill either. I never feel like I am participating in a living, breathing, tabletop campaign world when I am playing a CRPG. So what is missing?

To some degree, what is missing is smoke and mirrors. When I GM, I don't keep close tabs on every NPC, but when it comes time for a player to interact with that PC, it seems like I have done just that. Likewise, I am not really keeping track of iron ore supplies in Albert's kingdom, but when it comes time for Albert to go to war, I will have listed a shortage of iron ore a reason and made sure there now IS a shortage with the proper effects.

I suppose that my thought is, what about instead of tracking 100s of NPCs with sophisticated AI and adjudicating the interactions between those AIs, why not use a mix of scripted and procedural content to create a single system with the ability (I wouldn't exactly call it AI) to "GM a good game"?

I am imagining

-- A core engine that deals with the game rules, adjudicates player actions, and interacts with the UI.
-- A GM component that has the job of giving the player something reasonable to do as the game goes on. The GM component knows what quest/event scripts exist and picks between them based on the player's situation. Once a quest script is active, the GM component has to coordinate calls to the script as the quest progresses. (a little fuzzy here, sorry)
-- A script services component that exposes a set of procedures that generate dynamic content for the game and primarily the quest scripts. So, one routine might take a set of parameters and generate a random dungeon. Another might create an enemy encampment.
-- A conversation engine that uses a combination of traditional conversation trees, character skill use, character "history", and relationship tracking to create meaningful NPC conversations (that is for another post)
-- Event and Quest scripts using a well-known scripting language like javaScript, VBScript, or Python. These scripts would be required to have a specific "interface" (QuestRegister, QuestInitialize, QuestStart, QuestStateChanged, QuestCompleted... that sort of thing) that the GM component could call into and would call the GM and services component to access the game state.

So, our test player is walking around the game world and the GM component decides it would be a good idea to run some kind of event since there have been no pending events or quests for a few days. It looks at all of the registered quests and cross-references the player's current location, situation, faction standings, etc. From this, it generates a list of possible events and picks one randomly from that list.

For simplicity's sake, let's say the GM component comes up with quest#248 which is lovingly titled, "Yet Another Camp of Goblins Pillaging the Countryside". From the meta-data it has for this quest, the GM component knows to initialize the quest once the PC enters the next small settlement.

The PC does in fact enter the village of Helplessville and the GM sees our quest#248.Initialize message on its list of events to process. Thus, the GM calls into the script and runs the initialize routine. This routine calls into the services component to generate an enemy encampment, and an enemy boss (a goblin of level PC+1). It then looks at all of the inhabitants of the settlement and if they are of a low rank, it writes a (pre-written) conversation tree into the NPC's existing tree that largely has a few rumors and rumblings about the goblin attacks, and perhaps a suggestion to see X (where X is the highest ranked inhabitant) if the player wants to help. If the inhabitant is the highest ranking NPC in the village, it writes another conversation tree that has details about the goblin attacks, including the name of the lead goblin (has to be a way for the conversation tree to account for such variables) and where the encampment is.

Similarly, when the player talks to village headman X, the GM component knows to call quest#248.Start which will run more code based on what the quest author envisions happening. When the PC finally gets to the location of the encampment, quest#248.StatusChange runs, and so on.

Now the use of a full-featured script language gives us the ability to do Daggerfall one better (scary that we are still talking about Daggerfall in 2006) in that we can have a pretty complex, branching, randomized (if we like) quest that will allow the GM component to resuse the same quest script, but with different permutations each time. For example, what if when our hero reaches the encampment, a random number is generated and the quest branches from there. So, the PC finds the goblins are all killed. He could just go back and collect his reward, or he could explore the encampment and battle the beast that did this. Another branch might be that the PC finds that the culprits are not goblins at all, but humans in the employ of Lord UsurperDude who has designs on this land.

Now, this could get painfully comlplicated, for certain. If our randomized, branching quest comes up with the Lord UsurperDude option, then the quest programmer has to make sure he covers the options that now exist. What if the PC returns with this information? What if he kills the Lord's men? What if he does not complete the quest, but instead goes to talk to Lord UsurperDude? This kind of interaction is possible with this system, but the coding of a complex quest would be a non-trivial task.

In any case, let's say we end up with a very simple option. Namely, there are bad goblins here that must be dealt with, and I will simplify even further, and say that the quest writer did not make arrangements for a diplomatic option. (Given that he procedurally generated the boss goblin, I suppose he would have had to have added a pre-written conversation tree to the goblin to allow diplomacy. Perhaps that would be another branch option?) The PC defeats the goblins, but let's say that the leader flees and escapes the combat.

Great, so quest completed, the PC returns to headman X and gets his reward. The quest complete routine in the script is called for this. The PC's reputation with this village increases. His rep with the goblins decreases. He gets some cash and experience. But to keep with the concept of an illusion of a living world, some representation of the history of this event is written for later reference. A record of the goblin boss is written, as is his current status (alive, location unknown).

Later down the road, when the player triggers another quest, the quest author might invoke the service routine FindMeARecurringVillain. The routine chooses our goblin boss who is then leveled up appropriately, given some fluff dialog about "remembering how you ruined my plans at Helplessville" and placed in the new quest.

So what we have is a semi-emergent model. An expert GM system chooses relevant scripted content, and that scripted content uses procedurally generated content to fill in many of the details.

I hope this has been interesting enough to warrant posting it in this excellent thread,

 

[kingcomrade]

So, basically, you are simply talking about applying the way Diablo generates dungeons to quests? Having a couple dozen pre-defined (handmade) pieces, and then having the game patch them together?

 

[Oarfish]

cmagoun, It sounds to me like you are simply moving the problem into the 'GM' component and expecting it to have intelligence. For a system to procedurally generate narrative it will still have to use some kind of simulation or manipulate more abstract representation of 'plot'. Reducing the fidelity of that simulation because most of it isn't required for a single path through the game isn't going to solve the problem. Though I do agree that burning god knows how many cycles managing the intricacies of rope weaving, potty training and furtive masturbation might be getting away from the problem at hand - creating a believable, reactive game world that allows the player to do interesting things.

Differing levels of simulation fidelity depending on the proximity of the player might be an approach to reducing the amount of processing power required, but that will still bring up problems passing the results from one level of the simulation to the next.

 

[cmagoun]

cmagoun, It sounds to me like you are simply moving the problem into the 'GM' component and expecting it to have intelligence. For a system to procedurally generate narrative it will still have to use some kind of simulation or manipulate more abstract representation of 'plot'.

Conceptually, I am moving some of the processing to the GM component, as opposed to dealing with the interaction of 100 NPCs. However, I don't see the GM as needing to be particularly intelligent (and perhaps GM is a loaded term that is misleading). The GM component has the job of deciding when to run a plot, choosing from the possible scripted plots, and once a plot is "running", have some mechanism for deciding when to call that plot script's various routines. So, it is more like a search engine coupled with an event handler/router.

The difficulty (and where I am fuzzy on the details) is how the GM component is going to hook these events. So, when the GM component decides it needs to initialize a quest "when the player enters a settlement", I need some programmatic way of determining when that happens. A lot of this deals with what properties the various entities in the game will ultimately have (for instance, if the player enters a zone, and that zone's type property == "small settlement") and how the GM component interacts with the core engine. I am not there yet.

As for plot, I am relying on the quest creator to write the plot. The quests and events are in fact, scripted. One difference is that the quest model is branching and if desired, randomized so that a single quest script might support a couple dozen permutations (and of course the engine and scripting language must support this).

The other difference, and where I thought it might be pertinent to this discussion, is the use of procedurally generated content to fill in the details and also, to make the situation more relevant to the player's experience. Filling in the details might mean generating random dungeons, treasures, enemies and so on. Making the situation more pertinent means keeping track of NPCs that are relevant to the player and using those "intelligently" to take the place of actors in the quest script.

So we get a smoke and mirrors approach. King Albert does not go to war because his AI says so. The GM/quest script decides that a war starts, and we pick King Albert because he and the PC have a high relationship score.

Sorry to be unclear,

 

[Oarfish]

Ah, think I get you now. It's more likely my reading skills than your explanation.

I big problem I can see with integrating procedural and man made content is fragility - a human made quest may easily be broken by doing something unexpected if the world is not purely dictated by the author. This is assuming the procedural game world is truly interactive and not just Bethesdian "Clutter", re spawning guards and blood money payoffs.

I think I can see how this might work though, are you describing some sort of quest 'schema' where procedural content is queried based on its attributes and the actors integrated into the quest or created if needed? That seems like a pretty good idea, but still partly relies on having an interesting and detailed procedural world in the first place - if the static content is not going to have to account for everything of note.

 

[crufty]

Maybe an approach would be to fake it.

NPCs wouldn't have to figure stuff out, when generating a quest, you can also generate a solution. Finite combiations of quests = finite solutions.

To spawn quests, perhaps there would need to be not only locations at the player level, but an operational level map. Using hexes as a convention of scale, Lord Foozle raises his undead in his ruins in hex F10, with a ZOC of say 2 hexes. Just so happens there is a village in hex D9. The player is in hex A10, and hears a barkeep say "There's rumors that the dead are leaving their grave in the nearby village of Foozlekein."

So maybe the PC does something, maybe they don't. The end effect is, what drives the PC actions is an operational meta-game going on behind the scenes. Speech could be procedurally driven babelfish style; basic concepts fed through translators.

Complex stuff. But not impossible. Fun though?? *shrug*

 

[cmagoun]

I think I can see how this might work though, are you describing some sort of quest 'schema' where procedural content is queried based on its attributes and the actors integrated into the quest or created if needed? That seems like a pretty good idea, but still partly relies on having an interesting and detailed procedural world in the first place - if the static content is not going to have to account for everything of note.

Yes, this is a pretty succinct way of saying it -- much clearer than I did, really.

As to your other comments:

I big problem I can see with integrating procedural and man made content is fragility - a human made quest may easily be broken by doing something unexpected if the world is not purely dictated by the author. This is assuming the procedural game world is truly interactive and not just Bethesdian "Clutter", re spawning guards and blood money payoffs.

This would be a reasonably easy system to implement, but very hard to imlpement well. If you do have branched/randomized quests, you are relying on your quest scriptors to have a very good idea of the implications and effects of any of the branches on the game world. It would be easy to write a quest that had 100 permutations, but was buggy as hell. Add to that your concern that static and procedural content will clash, and yeah, you have a potential mess.

There are some things you can do to mitigate this mess though. If your services component is written well, when a script calls for GiveMeFriendlyNPCReadyToBeVictimizedByTheBadGuys, the system knows not to return an NPC that is already "in use" by any other content (static or procedural). Similarly, a procedure that returns a dungeon should be able to give a location for that dungeon that does not overlap some other feature, or is in an unreasonable position (such as 3 feet from town). Unfortunately, it also means there needs to be a way that a content creator can mark an NPC, or location as "used" and that content that absolutely requires a certain NPC will have to be able to register that somewhere with the GM.

 

[galsiah]

cmagoun:
I think it's worth looking at a more player centred approach - that is how P&P games usually work, as you say. I also think it'd be very hard to make convincing.

For instance, does your "GM" scale encounters to the player's ability? If you're not careful, things will seem as unconvincing as Oblivion. If you simulate the world, then let the player interact with it, it will automatically be cohesive. If you design the world around the player, then you need to do it very well to avoid it seeming contrived.

One of the reasons a good GM can do things well from a player centred persepctive is that he can improvise where necessary - even if he hadn't considered a problem was possible before. An expert system can't do this - unless the problem fits into its framework.

Another task that's fairly easy (ish) for a GM is to backtrack the implications of a largescale decision. For instance, say it is decided that country X has an iron shortage, so needs to go to war - for plot purposes. What effects should this have had on that region or others? Is there an explanation for the causes of the shortage? What knock-on effects would that have? Have iron prices been rising recently? If not, why not?

A GM can decide to make an instant change (e.g. national Iron shortage) that would in fact have had been building up in that region for a long time. It might be hard to make it convincing and explain things, but he can usually do it if he's clever - of course if he can't think of a reasonable explanation, he won't have it happen.

An AI GM is going to have a load of trouble with things like this. It's probably possible, but I'd guess doing it robustly and reliably would be harder than simulating the world in the first place. You'd have to change the state of the world now, extrapolate back in time to find what the situation would have been like to lead to the current state, then extrapolate forward from there to see how the current state should change - making sure to be consistent with the player's observations at all times.

If you don't do things like that (which would be a nightmare), you'd need to do things how a human GM would do it - by coming up with possible causes of the situation, then quickly guessing at the implications of these causes on the current state of the world, then making appropriate alterations when the players come across areas of change. I doubt that this would be practical - it's just not easy to get a computer GM to separate the "obviously important" consequences from inconsequential details.

You could avoid this by simulating everything forward in time - i.e. at a certain point, have region X's iron supply run out, and let things go on from there. However, this would have nothing like the immediacy of a human DM - the war might not start for a month, and the player might be in a completely different situation by then.

I think simulating the world, then having the player interact with it is probably the better solution, since the world will always be cohesive, and events will always make sense automatically. Creating a world that makes sense from a player centred view, and creating quests from thin air would be much harder to make convincing.
I also think that it'd be hard to get enough premade quest templates so that things didn't get repetitive.

If this were tried I'd suggest not doing the following though:

when a script calls for GiveMeFriendlyNPCReadyToBeVictimizedByTheBadGuys, the system knows not to return an NPC that is already "in use" by any other content (static or procedural).

While you might want to protect static, hand crafted NPCs, I think that it would be best (if at all possible) to allow interactions between procedural quests. To allow this, you'd need to make sure that if NPC X is killed at any time, the quest responds appropriately - you'd probably need to do this in any case, since the player might kill him at any time.
This might mean that <useful NPC X> is selected as <victim for quest Y>, and ends up dead. This might well change things in the first quest, but it makes things more interesting. The more interactions you have, the more situations will seem real, and not scripted. Individual quests would need to be designed robustly, but that's preferable in any case - if you are planning to show the same content to the player a few times, it is quite likely that he'll do something unexpected at some point.

If some quests are occasionally made impossible after an important NPC dies, or another interaction changes things (e.g. two countries starting a war), that's ok. So long as the quest fails reasonably, rather than breaking strangely.
When a quest "breaks", the main problem is usually not that it becomes impossible, but that it does this senselessly. If you've got a potentially infinite number of proceedurally placed quests, one of them ending prematurely is not a disaster, so long as this happens elegantly and convincingly.

 

[Allanon]

So long as you subdivide things well, there is no trouble with having the potential for a merchant to answer 100 questions.

But do you realize the complexity of the data base behind this massive world? With each NPC supposedly "living" a life, there will be huge amount of information to track.
Look here:

I might use disturbances in wildlife to give me clues.

You really wish the game to simulate the world to such a degree? What you are describing here is a system that simulates a very high number of real life aspects. Again, since you seem to understand the technical side of this better than me, do you really believe this is practical?

That doesn't make occasional failure a bad thing.

Never said it is. However, the system you describe has high chances of generating unsolvable quests. What bothers me the most is the influence of those quests on the gameplay. When the process itself is rewarding failure might be an option, but it's not always so. Would it possible to program it in a way to avoid such pitfalls, but not to interfere with the integrity of the world? I assume that if that would be possible, than adding designed quests should also be possible.

Your argument might have some force if RPG designers created "arbitrary" triggers, but that hasn't been my experience. If you have examples, feel free to share.

That's not me.

If you can't achieve something, it is always because the world is like that - not because a designer didn't think you'd do things in the way you did.

Again, only in case you simulate the world to such degree.

There being a lot of possible interactions doesn't make a difference if most NPCs don't use 95% of them most of the time - and they wouldn't.

Why? You are describing a system where even npc your character has never met might leave traces in the woods...

The player can easily refuse to help if a task seems boring.

That might mean going through tens of different quest givers before finding something worthwhile? That's how you imagine a crpg from a gameplay point of view? I'm not saying you will necessarily have to do that, but the potential is there. That's why I strongly believe that you must have pre designed quests side by side with any dynamic content.

So what we have is a semi-emergent model. An expert GM system chooses relevant scripted content, and that scripted content uses procedurally generated content to fill in many of the details.

I really like that implementation. It has the potential to unite both well designed quests with random content. Not sure about the technical aspects though...seems like your "GM" might have problems predicting the implications of players actions.

 

[galsiah]

That's not me.

Well... it sounded like you... close enough anyway .

But do you realize the complexity of the data base behind this massive world? With each NPC supposedly "living" a life, there will be huge amount of information to track.

Sure - it's complex. I do wish people would stop viewing complexity as some sort of disease, or equating it with impossibility.

The complexity just means you need a good design which organises things well, so that the complexity is managed. In any case, a list of responses which is 1000 entries long is not really much more complex than a list 10 entries long. For most purposes, both can be treated as "A list of responses". The 1000 entry list will probably have a lot more connections - so would need to be organised and divided etc. However, the simple fact that there is a lot of information, is not a problem - it just means you need good organisation.

Can I give you a wonderfully organised system now in five minutes? No - it'd take some design work. That doesn't make it impossible though. Complexity is just an issue to be aware of, and to deal with.

I might use disturbances in wildlife to give me clues.

You really wish the game to simulate the world to such a degree? What you are describing here is a system that simulates a very high number of real life aspects. Again, since you seem to understand the technical side of this better than me, do you really believe this is practical?

(1) This is was not necessarily intended as an example of an emergent quest - just as an illustration that enjoyment can come from a process, rather than successful completion.

(2) It probably is reasonable, since:
You are specifically following the NPC, and he is running from you, so this is not just a random NPC, but an NPC the game could reasonably treat as a special case automatically. Could a computer cope with calculating tracks for the few NPCs that are close to the player? Certainly.
By "disturbances in wildlife", I just mean: "Look, that dear is running off to the west - perhaps it was scared off by the guy I'm tracking. I'm probably in the right area."

Again, nothing in this needs to be done specifically as a "quest". Deer and birds etc. can automatically run/fly away from disturbances; tracks can always be calculated for NPCs outside towns and within X distance of the player; thieves can always try to escape from any character they think is pursuing them.

All these things are not special extras that occur to generate quests, but are just parts of the game world which happen to influence your current goals. There is no need for quest scripts (not that they exist) to know about the thief, the deer, the tracks etc. The thief and deer act independently, and the tracks can always appear in the wilderness for NPCs near the player.

Never said it is. However, the system you describe has high chances of generating unsolvable quests.

Why? Please explain. Don't assume that the first, naive implementation would be used - why would it not be possible to adjust things so that failure (or at least unpredicatble impossibility) were very rare?

What bothers me the most is the influence of those quests on the gameplay. When the process itself is rewarding failure might be an option, but it's not always so. Would it possible to program it in a way to avoid such pitfalls, but not to interfere with the integrity of the world?

Hopefully. Personally, I'd expect that failure in emergent "quests" would usually be more interesting than failure in hand crafted quests. In emergent situations, you are free to try any tactic, and can feel that what you've tried and the results you get are all down to your specific methods - not because you followed the script from A to B to C. Even when you don't succeed, the methods you used would have been your own.

When you fail at a scripted quest, it is usually because you are stuck on a stage because you don't know what you are supposed to do next. When you try various reasonable approaches (some of which really should have worked), but fail, you don't feel like you're getting closer, but rather you feel frustrated.

Failing at an emergent quest is more like (again, sorry ) failing to bring down a retreating enemy with a bow. Perhaps your methods are clumsy (you miss half the time), but succeed; perhaps your methods are clever and precise (you take up the best position, and hit accurately many times). Even if the enemy gets away, you can feel satisfied if you did the most that you could very accurately, but it escaped - and if the escape is reasonable.

In the same way, you can feel very satisfied by your skill or intelligence when trying an emergent quest - even if you don't succeed. Perhaps the quest is to rid the area of some powerful creature, but it is very well armored (so the situation actually is as above). If you hit it with five arrows in quick succession, but have no way to bring it down before it hides and heals, you'll fail - but you'll feel like a damn good archer. Perhaps you need to find the NPC who stole from a certain secret chamber on a certain night - you might narrow it down to seven NPCs who had access, then eliminate five by careful comparison of answers to questioning of other NPCs. Even if you can't find which of the last two was the thief, you can feel good about making progress.

I agree that it wouldn't be good for these failures to be very common, but I think things could be designed to achieve this.

I assume that if that would be possible, than adding designed quests should also be possible.

Yes - I'm not suggesting doing away with hand crafted quests. I'm just suggesting replacing filler quests with an emergent, NPC need based system. I'm mainly thinking of games like Morrowind or Oblivion, where there are many relatively insignificant NPCs. The main characters would need to remain hand crafted - presuming there are some. Presuming that there was a main plot / major plot lines, these would probably need to be hand made too.

If you can't achieve something, it is always because the world is like that - not because a designer didn't think you'd do things in the way you did.

Again, only in case you simulate the world to such degree.

This will be true however detailed the simulation. A more detailed simulation just makes things more interesting. I agree that such a simulation is non-trivial, and would need to be quite complex to create interesting gameplay. That's not impossible though - just hard.

There being a lot of possible interactions doesn't make a difference if most NPCs don't use 95% of them most of the time - and they wouldn't.

Why? You are describing a system where even npc your character has never met might leave traces in the woods...

See above re: "traces in the woods".

Most NPCs wouldn't use most options most of the time, simply because they are not relevant. For instance, every NPC would have a wide range of possible responses to player questioning, but these are very rarely used - only when the player asks the relevant question. Each NPC might have a wide variety of possible combat actions, but would completely ignore them when not in combat. An NPC might have access to complex algorithms to find food, but might not use them at all if the last solution they used is easy and still works.

Most of the time an NPC doesn't need to think anyway - he just continues what he is doing. If nothing unusual happens, he might only need to re-consider his actions every 30 seconds or so - perhaps less often for NPCs a long way from the player.

There might be level of detail issues with this - i.e. differences in the overall outcome of events depending on the player's presence. This is unlikely to be too much of a problem, since things being most active around the player is generally good.
However, it could lead to a situation where the player can travel to a town, and get odd answers when he asks what happened yesterday. This could be handled, I guess, but it might not be simple.

That might mean going through tens of different quest givers before finding something worthwhile? That's how you imagine a crpg from a gameplay point of view? I'm not saying you will necessarily have to do that, but the potential is there.

Sure it is, but most of the time the player will be able to find interesting tasks (hopefully) by asking in the right places. If you ask a peasant farmer if he needs help with anything, you usually won't expect to get an epic quest involving the clash of huge armies and the fall of civilizations - more likely something with potatoes.

If you ask a guild-master / town mayor / guard captain / bandit leader..., you're more likely to get something more influencial (though it won't necessarily always be interesting).

It's possible that the player could be given the option to ask for more difficult tasks, by responding to requests with e.g.:
"I'm sure you can handle that without my help - give me a challenge."
"That's a bit ambitious for my liking - perhaps you need help with something simpler."
"Is there anything I can do for you today - I'm on a tight schedule."
And so on.

An NPC (or family / faction / town / country...) could have needs prioritized according to importance and likely difficulty. So if a certain need is very important to the NPC, but there are three reasons he can't satisfy it, it would get a high priority and difficulty. If a need is unimportant, it would get a low priority. If a need only has one obstacle, it would get a low difficulty (obstacles could be prioritized too - e.g., "it's guarded by a dragon" is probably a bigger obstacle than "it's too far for me to walk").

An NPC would then make requests according to the player's responses, but would favour high priority tasks where possible (or perhaps even refuse to hand out other tasks if priorities of some were extremely high - e.g. a starving peasant might only hand out "please get me some food" even if the player wants something hard, and the NPC also has "please bring the gang who stole my food to justice").

Again, I agree that this would all get rather complicated. To do this well would be complicated. It might be worth it too.

That's why I strongly believe that you must have pre designed quests side by side with any dynamic content.

I agree - I'm not suggesting a 100% emergent quest system.(not if you're aiming to make a conventional RPG in any case)

So what we have is a semi-emergent model. An expert GM system chooses relevant scripted content, and that scripted content uses procedurally generated content to fill in many of the details.

I really like that implementation...

It's more of an idea than an implementation. I think it'd need to be a bit more fleshed out / prototyped to be called an implementation.

It's an interesting idea, I agree, but I still think it'd be even harder to do well than a complete simulation with "quests" arrising from simulated needs (and that isn't easy either).


Section8:
Lots of nice ideas. I particularly like the idea of moderating stock NPC dialogue responses according to needs. It'd be nice if needs based "quests" could emerge relatively naturally in dialogue (as far as possible) rather than through the player asking directly for a list of needs. Again, it's probably hard to do well, but it could be very effective.

 

[Section8]

There's a lot of ideas coming across here, and rather than addressing specific quoted fragments, I'll just generalise a bit.

Human GMs vs AI GMs

I think the the overall concept here is already a similar model of the role/duty of a human GM, with the only big difference being that the human tends to work top-down when it comes to generating appropriate content, while the AI does the opposite.

Think about it. When you're GMing, you'll no doubt have "background processes" running for everything the players might potentially interact with. The NPCs in a far off land the players may visit one day, all have a bare bones "personality" based on the culture of the place they live. The player may be adventuring in the high fantasy derivative of Feudal Japan, but somewhere across the sea, and in the back of the GMs mind, the Chivalric Medieval Europe derived lands are full of knights in shining armour and what not.

It's when the players begin to interact, or begin to receive foreshadowing narrative that details come to the fore. Last weeks session left off with the party just passing through the gate of the capital city, and so this weeks prep is likely to include a lot of notes on the city districts, important NPCs, shops, etc.

As a human, that has to be a bit selective. If my city has ten blacksmiths, I might only work up three individual personalities and apply them according to where the players choose to go. <X> personality will go to Bob the Blacksmith, just because he's the first one the player chose. And that's how I see cmagoun's idea. You've got scripts to be applied on demand, with enough variables to keep things distinct.

But that to me seems like you're missing out on the big advantage a CPU provides over a human brain. The human has to cut corners on content and apply on demand because it's impossible to account for everything the players might choose to do. A program isn't limited in the same fashion. It can simulate everything locally, and do as the player does for distant affairs, and keep them as background processes, but even in this regard it can track far-off settlements and calculate a simple product of the collective systems at play.

Narrative / Mixed Procedural & Scripted Content

And that leads nicely into the next idea. If an actual storyline, detailed characterisations and other various narrative elements are required, then cmagoun's idea gels better with mixed content.

But for me personally, I don't think strong narrative elements are necessary. I've never really enjoyed chasing or being forced along some designer's idea of an epic story when contrasted against having actual freedom. I don't mind chasing a narrative in adventure games and such, because that's just the nature of the beast. Not so with RPGs.

And so, I'd push for 100% procedural content. That way, there will never be any conflict between the robust "anything can happen" of emergent gameplay and the fragility of limited scripting. With enough scope for "dramatic" events, such as armies amassing and capturing cities and such like, I think the "story" would comfortably write itself, even if it was fairly bare-bones.

 

[Allanon]

That's not me.

Well... it sounded like you... close enough anyway .

You underestimate me. Not seeing the arbitrary triggers is fairly hard in modern days crpgs

Sure - it's complex. I do wish people would stop viewing complexity as some sort of disease, or equating it with impossibility.

Common, give me a break. I'm not saying it is impossible, or shouldn't be done just because it's complex. However, unless you're planning a crpg with no sound, physics and graphics you need to take the strain on the hardware into consideration. Even if we take the flawed (imho) ideas of simulating mostly the local events (with less attention to farther places) you still come up with need for more cycles than the cpu can currently give. That's why it's always a good idea to question yourself if things can be done differently, in a more efficient way, without compromising the great results.

You are specifically following the NPC, and he is running from you, so this is not just a random NPC, but an NPC the game could reasonably treat as a special case automatically.

Not really. At least the way I understand the system you describe. All actions done by npcs and therefore all the consequences are simulated as if the player character did not exist in the first place. That's how we intent to create a simulation of a real world, instead of smoke and mirrors. So, since the game can't know what quest the player will choose to pursue, it should simulate all of them with the same details. So, in our example, in a big city we might have tens of active thieves. At least some of them should leave traces. One thief might have sold the loot to the local sellers, other might have been seen by the night guards, another one might have fled to the forest, someone might have been murdered...the game will have to simulate all those aspects even before the player choose what thief he will be following. Moreover, take into consideration that all of those cases could have happened a week before the player even begun the quest...so it should continue to simulate the consequences and actions for at least so long. In the end it's just a huge amount of data.

Don't assume that the first, naive implementation would be used - why would it not be possible to adjust things so that failure (or at least unpredicatble impossibility) were very rare?

Because the minute you adjust it in that way, you pretty much screw up the integrity. The gaming world will no longer live by a set of standard rules, but will abide some artificial guidelines to make it more playable. For example, a good thief with that system implemented might find he isn’t so good - his perfect crimes might be not so perfect some of the time, even though abilities wise he is capable of making them always. Why? So that the player will have higher chances of avoiding dead ends. That might as well mean that in a specific case that thief might be caught by the guards and executed (as a result of tempering with his skills).

Even when you don't succeed, the methods you used would have been your own.

Yes, indeed. However, how exactly player own skills will compare to the npcs? So far we assumed the player can do pretty much what he wants. Since the world is simulated in such way as if the player didn't exist, npcs will have their abilities and skills relative to each other. The player's skills and abilities might not be on par with the requirements for certain quests. So, the player will have to wander around and filter quests based on their difficulty. That is not new, since we are quite used to that in role playing games. However, unlike crpgs we play today, procedurally generated worlds will have two kinds of unsolvable quests - those that are too difficult and those that are dead ends. I believe this might have quite an impact on the gameplay. In Oblivion the designers tried to solve similar problem by adjusting the world to the player, but it didn't go well. On one hand you can take any quest any time, on the other hand it takes out the challenge.

The main characters would need to remain hand crafted - presuming there are some.

The problem is that unless you want them to look like puppets you'll have to make them live believably in the world. That might cause you to loose control. How would you guard them from being killed or distracted or wandering away....

However, it could lead to a situation where the player can travel to a town, and get odd answers when he asks what happened yesterday. This could be handled, I guess, but it might not be simple.

My thoughts exactly. It might screw everything up.

It's an interesting idea, I agree, but I still think it'd be even harder to do well than a complete simulation with "quests" arrising from simulated needs (and that isn't easy either).

Frankly, right now both of them seem theoretical at best.

It can simulate everything locally, and do as the player does for distant affairs, and keep them as background processes, but even in this regard it can track far-off settlements and calculate a simple product of the collective systems at play.

I don't think it will work well. Locally you won't feel any difference, but on the bigger scale, it will be noticeable that distant affairs aren't calculated. Consider two neighboring towns in a state of conflict over a gold mine on the borderline. While the player is in the first town, the AI will actively simulate all npcs, causing the mayor to decide on a sneak attack. If both towns were calculated with the same details, the second town might have spies in the woods that would alert about the attack...
Somewhat lame example, but I hope you get it

 

[Zomg]

Common, give me a break. I'm not saying it is impossible, or shouldn't be done just because it's complex. However, unless you're planning a crpg with no sound, physics and graphics you need to take the strain on the hardware into consideration. Even if we take the flawed (imho) ideas of simulating mostly the local events (with less attention to farther places) you still come up with need for more cycles than the cpu can currently give.

That's rank speculation. I'm not saying it isn't true, but no one in this thread has done the math on the memory and processor load caused by a lot of integer manipulations of many arrays with "need" elements. I would say in favor of this needs stuff that it would be highly parallelizable, meaning it would be easy to implement on the multiple CPU systems that seems to be the future. There is also no presumed "base world" in this discussion - maybe this would have to be done in a Roguelike state->state world rather than a RT one.

I'll agree with the 100% procedural-emergent solution myself. I think eventually you could come up with some tricksies to integrate HM and P/E, like modular characters (say, a farmer character that can incarnate in any particular farmer you run across) and similar stuff, but even failing that if you want a unifying narrative you could take the Darklands route (God comes to you in a dream).

 

[crufty]

Ok, so procedural content is the way to go. Lets take a step back..before the game...

What algorithms are there to make procedural landscapes that look real? W/forests in mountains, shrubs in lowlands and palm trees on the beach? farms? cities? villages, complete with crooked ass village streets and mindless farmers sitting on rocking chairs?

 

[cmagoun]

cmagoun:
I think it's worth looking at a more player centred approach - that is how P&P games usually work, as you say. I also think it'd be very hard to make convincing.

For instance, does your "GM" scale encounters to the player's ability? If you're not careful, things will seem as unconvincing as Oblivion. If you simulate the world, then let the player interact with it, it will automatically be cohesive. If you design the world around the player, then you need to do it very well to avoid it seeming contrived.

To my mind, the solution is a mix of static content, scaled procedural content, non-scaled procedural content, and guided procedural content. You might walk over the mountain and find a citadel full of 25th level creatures that will kill you outright. On the other hand, some encounters might be scaled to be the "correct" level of challenge and still others will be "sort of" scaled within a range set by the writer of the encounter. Frankly, this does not seem hard to do, and it is more likely that BethSoft was lazy in its design thanit is proper "scaling" is a monumental task.

One of the reasons a good GM can do things well from a player centred persepctive is that he can improvise where necessary - even if he hadn't considered a problem was possible before. An expert system can't do this - unless the problem fits into its framework.

Yes, but this problem exists no matter where you put the processing. AI is a problem that has not been solved for 40 years. The 100 NPCs running around with their needs, motivations and actions will only be able to interact in ways that are pre-designed into their frameworks. And while there is something to be said about multiple simple interactions building on one another to become complex (the whole universe is based on this fact), I am not sure that it holds in this case. It might be possible that 100 entities, having only access to fairly basic interactions will come together to form a complex story, it is just as likely you will get lots of banal stories.

Having said that, I understand your concern, and you have a good point, but I feel that it exists with any of the systems we would discuss in this thread.

Another task that's fairly easy (ish) for a GM is to backtrack the implications of a largescale decision. For instance, say it is decided that country X has an iron shortage, so needs to go to war - for plot purposes. What effects should this have had on that region or others? Is there an explanation for the causes of the shortage? What knock-on effects would that have? Have iron prices been rising recently? If not, why not?

A GM can decide to make an instant change (e.g. national Iron shortage) that would in fact have had been building up in that region for a long time. It might be hard to make it convincing and explain things, but he can usually do it if he's clever - of course if he can't think of a reasonable explanation, he won't have it happen.

An AI GM is going to have a load of trouble with things like this...

Again, I regret calling any of the pieces of my idea "GM", since I think it is a misleading term. Most of the intelligence in the system is really centered in the quests, like in a traditional scripted system. So, if I wanted to write an event script that would start a war between two neighboring nations, I might write a script that has this flow to it:

--Find two nations that border one another (preferably ones that the player has ties to, or is near)
--Pick a reason to go to war from the 6 I have thought about (I being the quest writer, not the AI)

--If the reason is "Resource Shortage"
----Find a resource that both nations have (this implies that we track a resources list for a nation object)
----Pick one nation to have a shortage
----Drastically raise prices in the shortage nation (of the resource and anything based on that resource) OR Apply a marker to that nation that will dynamically increase prices when the player looks to buy/sell something based on that resource (easier to do).
----Increase the chance of the WarTimeEncounters script to run in the two nations (presumably, I have thought of potential encounters that might happen in a nation at war such as bandits, profiteers, soldiers, refugees as well as events such as a peace being called)
----Initialize the WarTimeQuest scripts for nobles in both countries (that would give the PC war-time missions such as to hunt down enemy soldiers, spy on their camps, etc.)

--If the reason is "Romeo and Juiet-Like Scenario"
----Find a male heir in one nation, female in the other of appropriate age, or create them.
----Add dialog tree "My Son is missing and I blame Nation A" to Nation B
----Add dialog tree "My Daughter is missing and I blame Nation B" to Nation A...

...and so on...

All of this is programmed by a human into a non-compiled script that can later be tweaked, modified and added to by others. This is a mix of procedural and traditional, but the onus is on the quest programmer and the programmer of the "service component" that finds or creates appropriate content to fill in the details. The GM component is really just a layer that decides what quest programs to run at what times.

I think simulating the world, then having the player interact with it is probably the better solution, since the world will always be cohesive, and events will always make sense automatically. Creating a world that makes sense from a player centred view, and creating quests from thin air would be much harder to make convincing.

I grant you this, it is by far the more interesting solution, but even if we could manage to do it successfully, would it make a playable game? I mean, we do have a very good example of a world simulator going right now and yes, it features cause and effect and does create very interesting situations. However, for the vast majority of the actors in this simulation, the interesting stuff is far removed from them. How many people's lives make interesting novels?

(And for the sake of future discussions, we are assuming the answers to that question is yes we can do it and yes it would make a good game because it is interesting to talk about. I suppose my point is that you are going to have to contrive and scale even the best world simulation just to make sure that the player can effectively participate in events.)

I also think that it'd be hard to get enough premade quest templates so that things didn't get repetitive.

I agree that this is a problem. Keep in mind that from a practical perspective, we only need 100 hours of quest templates to put most commercial RPGs to shame. From an idealist perspective, a little repetition is not necessarily a bad thing. I didn't mind going to the a guild in Daggerfall and getting yet another "get object X from dungeon Y" quest. What I did mind is that it was NEVER any different.

I would have loved to see is a system that every so often, threw me a curve. So, I get a couple quests done and am on my third "get x from y" and I am getting a little bored with it. However, when I get there, I don't find x. Instead, I find bloody bodies lying all over the place. Item x is gone, but there is a note.

"Dear InsertYourNameHere,

You have annoyed us in the past, and now you are going to pay for your transgressions. If you want to find item x, you will have to come to the Randy Badger tavern in SomeTown by the 5th Sunday in May. We will be waiting. Bring some hefty cash.

Sincerely,
The Guild You Recently Defeated In A Previous Quest"

Now, I have to raise some cash quickly, and I do so, but when I get there...

--It could be a trap and there could be a huge ambush waiting for me
--It could be legit and though there are a lot of baddies there, they will allow me to make the exchange
--It could be a bluff and the evil guild has returned the item to my patron and told him I intended to make off with x and they retrieved it (leading to complications when I return to report my failure)
--It could be a bluff and they have the item back at their base (which I now have to infiltrate)

That is a game that would hook me. All you need to make a generic quest into a memorable experience is for some twist to happen some small percentage of the time. I would do the repetitive quest with the normal rewards, knowing that there is a chance of something interesting happening every time. Even in the "template" described, you have 4 possible permutations (and this storyline is merely a permuation of the main get x from y template). Make 50 such templates with an average of 6 permutations each and you have a pretty varied experience... at least out of the box.

Thanks for discussing this with me,

 

[muds_animal_friend]

I don't think it makes sense to call an emergent system based on NPC needs a "mere labour saving device".
[emp. mine]

Neither do I, which is why I didn't. I called content generation systems mere labour saving devices, which they are. They're labour saving devices because there's nothing such systems can generate that can't be created statically by human authors (albeit at a greater cost in time).

Dynamic content generation and emergent gameplay/narrative aren’t mutually inclusive. Even in your NPC needs-based example a content generation system isn't required; all the necessary data can be supplied by human authored content. In fact your example would probably be most viable when founded upon an "expert" or knowledge-based AI system with content generation reduced to – at most – stringing sentences chunks together.

 

[galsiah]

I think I'd certainly enjoy a 100% procedural game, but then I like strategy games. I think many fans of traditional RPGs would be put off by the absense of hand crafted story. Perhaps I'm mistaken.
If emergent story did turn out to be as compelling as (good) hand crafted story, then I'm definitely in favour of it - I'm just not sure that it would.

However, unless you're planning a crpg with no sound, physics and graphics you need to take the strain on the hardware into consideration. Even if we take the flawed (imho) ideas of simulating mostly the local events (with less attention to farther places) you still come up with need for more cycles than the cpu can currently give.

With respect, you have no idea whether this is true. If you tried to recalculate everything in the game world every frame, then you'd be right - so don't.
As I've said above, nearly every NPC needs to do pretty much zero processing 99% of the time.

You are specifically following the NPC, and he is running from you, so this is not just a random NPC, but an NPC the game could reasonably treat as a special case automatically.

Not really. At least the way I understand the system you describe. All actions done by npcs and therefore all the consequences are simulated as if the player character did not exist in the first place.

Just because I'm suggesting a complete simulation of the world, does not mean I'm suggesting doing it at the same level of detail when the player is miles away. There are always things which aren't calculated for far off parts of the world - the graphics and physics for a start.

NPCs would almost certainly be processed at a lower level of detail. This does not mean that fewer interaction types are possible - just that many of them might be estimated. For instance, if an NPC wanted to travel to X, Y and Z in his town during the day, it might just be assumed he'd made those trips and carried out appropriate actions. Whether he sees other NPCs during the day might just be simulated (perhaps at the time, perhaps only when the player asks), using say a 5% chance that he sees any of the other NPCs in the town. Various other reasonable estimates could be used for other things.

That's how we intent to create a simulation of a real world, instead of smoke and mirrors. So, since the game can't know what quest the player will choose to pursue, it should simulate all of them with the same details. So, in our example, in a big city we might have tens of active thieves. At least some of them should leave traces. One thief might have sold the loot to the local sellers, other might have been seen by the night guards, another one might have fled to the forest, someone might have been murdered...

First, what scale are you imagining? If we're talking on the scale of a city with 200 npcs, how many active thieves would there be? One? If you're talking of a realistically populated city, with perhaps 10000 NPCs, then of course not all of them will be processed exactly.
Again, just because I'm advocating a whole world simulation, does not mean that I'm saying it all needs to be done without estimation or fudging.

Perhaps it might not be possible to track every event at a high level of detail. My "persue the thief" example above was first not necessarily intended as an emergent quest example, and second was immediate - I wasn't thinking that my character was tracking the thief a week later, but that he was following him a few hundred yards behind.

the game will have to simulate all those aspects even before the player choose what thief he will be following. Moreover, take into consideration that all of those cases could have happened a week before the player even begun the quest...so it should continue to simulate the consequences and actions for at least so long. In the end it's just a huge amount of data.

Yes, there's a lot of data, but computers can handle a lot of data. You might need to simplify or estimate for areas a long way from the player. In some circumstances you could fill in the detail as he approaches. For instance, when the other side of the continent, processing theft might involve:
Thief A stole X from Y at time T, and was/wasn't caught. He then headed in direction D / was placed in prison P.

As the player nears that town, the simulation can check for recent events, and fill in the details - e.g. for a recent theft, the thief's exact route might be created, tracks might be placed, NPCs who might have been in the area can be told exactly where they were and what they saw...
So long as the simulation keeps track of the main pieces of information - i.e. the thief's identity, what was stolen, the time, where the thief is now, whether he was caught / seen..., there is no need to fill in the exact details until the player is in the area.

Again, this would not be easy to get working well, but it is possible for a simulation to keep tabs on events in the entire game world without grinding to a halt as it works out the texture of the footprints of thousands of NPCs.

Because the minute you adjust it in that way, you pretty much screw up the integrity. The gaming world will no longer live by a set of standard rules, but will abide some artificial guidelines to make it more playable.

Adjust it in which way? There are many different aspects to tweak, some of which involve compromise and approximation, while some don't. The design of the world is always artificial in any case. It might be hard to process things at different levels of detail, but still get similar overall results for a town regardless of the player's presence. Again - hard, not impossible.

For example, a good thief with that system implemented might find he isn’t so good - his perfect crimes might be not so perfect some of the time, even though abilities wise he is capable of making them always. Why? So that the player will have higher chances of avoiding dead ends. That might as well mean that in a specific case that thief might be caught by the guards and executed (as a result of tempering with his skills).

So don't do that. If having many masterful thieves around is going to screw things up, then don't - make most thieves in the world only somewhat skilled (probably more convincing in any case - master thieves are supposed to be rare).

You won't always be able to make things easy for the player - you won't want to either. So you need to make sure the player usually has an idea of the difficulty of the task he is taking on. Of course the game doesn't know how difficult it is, but that doesn't stop the player getting clues, for instance:

NPC has been robbed topic:
NPC: I've been robbed. <when and where it happened> <who saw the theft> <what was stolen> <information on the thief>

Situation (1):
NPC: I've been robbed. It happened in the market at noon, and half the town saw it. It was only a few gold coins, but that <NPC name> won't get away with it. He's clearly guilty - you can ask anyone. He ran off to the East, but he lives in the Southern slums. Perhaps you can track him down.

Situation (2):
NPC: I've been robbed. I'm not sure when it happened - I only noticed this morning. I don't think anyone saw anything of the thief. They stole a precious statue from a securely locked box, and nothing else. I can't tell you anything more. I'm sorry there's so little to go on. Perhaps you can track them down.

These dialogues can be built up using the NPC's knowledge of the theft. It should be fairly clear to the player that catching the first thief should be a simple matter, but finding the second might be impossible. The player can try to find the second thief, but if he gets nowhere, he shouldn't be that surprised - it was clearly going to be difficult.

After a while playing, the player should get a good idea of the sort of quests he'll be able to complete easily, and those which will be hard / impossible. After that point the challenge he sets himself is his decision.

So, the player will have to wander around and filter quests based on their difficulty. That is not new, since we are quite used to that in role playing games. However, unlike crpgs we play today, procedurally generated worlds will have two kinds of unsolvable quests - those that are too difficult and those that are dead ends. I believe this might have quite an impact on the gameplay.

There would rarely be any total dead ends - just quests that are too hard for one reason or another. Some quests might be almost impossible when you don't have enough information to go on - most would be only impractical though, not really impossible. That's a much more reasonable "dead end" than a dead end due to bad scripting though. There is some satisfaction in having done everything you can to achieve a goal, but running out of possibilities. There is no satisfaction in running into an artificial brick wall.

The problem is that unless you want them to look like puppets you'll have to make them live believably in the world. That might cause you to loose control. How would you guard them from being killed or distracted or wandering away...

That's not hard to do. You'd set up most NPCs to stay put in most circumstances in any case - half the merchants in a town shouldn't be wandering off. Also, combat should be very rare for most NPCs. A sensible game world would have most NPCs only resort to combat as a last resort. This means that you'd need reasonable town setups, with walls, guards, prisons etc., but that's not impossible.
Anyway, there's not much wrong if 1% of the time an important NPC does wander off - given that you can ask anyone when they last saw him (or perhaps what he was doing), it's unlikely he'd be that hard to track down.

Mostly you could control such NPCs within the general NPC framework anyway - just give them a strong desire to stay in a certain area.

My thoughts exactly. It might screw everything up.

Sure - no-one is saying all this would be easy. However, when something "screws everything up", there is very often a way to change things so that it doesn't.

It's an interesting idea, I agree, but I still think it'd be even harder to do well than a complete simulation with "quests" arrising from simulated needs (and that isn't easy either).

Frankly, right now both of them seem theoretical at best.

Absolutely - I never meant to imply otherwise. My ideas don't constitute a design or an implementation either. We're just throwing ideas around. Most of these methods should work in theory, but are sure to cause many difficulties in practice.

I don't think it will work well. Locally you won't feel any difference, but on the bigger scale, it will be noticeable that distant affairs aren't calculated.

This is certainly a difficult problem, but it can be tackled to some extent. Just because you are not simulating at the same level of detail does not mean you aren't covering all possibilities (or most of them at least). If it is possible for a town leader to "send out spies", you'd have this as an option for any town. The difference is that the actual events would be estimated, rather than carried out - e.g. the NPC spies would be placed in the woods rather than walking there, and once there they might have a percentage chance of detecting anyone within a certain radius...

Of course, it'd be nice if "sending out spies" could be an emergent, rather than preprogrammed tactic (in which case the action "send out spies" would not exist, so you couldn't simulate it at a higher level). This is pretty unlikely though, I think - at least until NPC AI is a few steps further forward.

There will always be differences when you simulate at different levels of detail for far off areas, but if you're careful to provide the same high level possibilities, those differences can be kept to a minimum (after a lot of design, testing, tweaking etc.).

 

[galsiah]

Yes, but this problem exists no matter where you put the processing. AI is a problem that has not been solved for 40 years. The 100 NPCs running around with their needs, motivations and actions will only be able to interact in ways that are pre-designed into their frameworks.

But these will automatically take any player action into account according to the rules of the game world. Scripted quests do not - they are only as versatile as their writers made them. This problem will only get worse if you try to apply the same quest template in different areas and different situations.
I'm not saying it's impossible to get working, but it is a difficulty which doesn't exist for needs based quests.

It might be possible that 100 entities, having only access to fairly basic interactions will come together to form a complex story, it is just as likely you will get lots of banal stories.

That's a much more reasonable criticism. I don't think any 100% emergent system is going to provide an overarching, compelling story from start to finish. Mainly I see it as a substitute for generic NPCs, or as the basis for a strategy-RPG hybrid.

It might be possible to get a decent story with 100% emergent quests, but scripted large scale events though. E.g. you script the causes of a war, predefine tendencies of actors and nations / other groups, including goals of groups and nations etc., then let the events drive the plot. You might get a good plot, but not that great a story like this.

...I might write a script that has this flow to it:
--Find two nations that border one another (preferably ones that the player has ties to, or is near)...
--If the reason is "Resource Shortage"
...
----Pick one nation to have a shortage
...

But my point was that resource shortages do not happen and cause wars in an instant - they build up over time with various knock on effects. How does the computer decide when it is reasonable to have a shortage of X? How does it make sure this fits reasonably with previous player observations?

For instance, if a player has travelled through a region a few days before a bought a few iron swords cheaply, then he hears about an iron shortage - or learns later that it occurred at this time - how can he make sense of this? Why were prices not already rising a few days before? Why were there no other indications as he travelled through that region - no escalated prices of other materials, no news of a shortage, no unrest...

To the player it is going to look like the shortage occurred instantly - which in fact it did. This won't seem reasonable.

You could get around this for this situation e.g. by making sure that the player hadn't been to that region for quite a while, or by triggering the start of the shortage, then having the effects emerge from there.

However, you'd have to fix these kind of problems for every new situation you script: If a goblin village suddenly appears, how do you explain the lack of any previous attacks from there, or the fact that merchants have been coming in that way safely for months.
If a cave entrance appears, why wasn't it there before? Do you only make scripted additions to areas the player has never been?

Again, these problems are probably not impossible to solve, but they'd be difficult to solve. If you use a bottom up world simulation approach, then all of these things are automatically explained (presuming that e.g. systems for placing new creatures are believable). Once you introduce level of detail solutions for world simulation - which would almost certainly be necessary -, things get harder, but the kind of problem you are dealing with is:
How come this thieves tracks have not been noticed or reported before now?
rather than
How come a national iron shortage has sprung up out of thin air without any warning signs or other consequences?

I grant you this, it is by far the more interesting solution, but even if we could manage to do it successfully, would it make a playable game?

This needs to be kept in mind, it's true - just because an emergent world simulation is theoretically interesting, does not necessarily imply that gameplay will be good. I think it could be though.

How many people's lives make interesting novels?

Your life would probably be more interesting if:
Death / disease / mutilation... only meant re-loading.
A friend's death allowed re-loading.
You didn't need to bother with most everyday chores.
Time spent in prison would pass instantly without aging.
You were placed in an interesting new world at the age of twenty.
There were large scale problems which could be tackled by brave heros.

I suppose my point is that you are going to have to contrive and scale even the best world simulation just to make sure that the player can effectively participate in events.

Sure - every game is contrived by definition. The initial world design will be contrived. An emergent world simulation is guaranteed to be cohesive though (according to its own rules), whereas procedurally added quests / areas might or might not fit well with things as they are.

I agree that this is a problem. Keep in mind that from a practical perspective, we only need 100 hours of quest templates to put most commercial RPGs to shame.

True, but I'd ideally like to put civilization to shame . Probably not practical for an RPG, but a 100% emergent RPG-strategy game has that potential.

From an idealist perspective, a little repetition is not necessarily a bad thing.... I would have loved to see is a system that every so often, threw me a curve....

Sure - I'd certainly enjoy that if it could be done well. That's one of the main reasons I think it's better not to make sure quests are kept separate (i.e. an NPC in one is never used simultaneously in another), but rather make them robust to changes. This would allow for things not to go as expected from time to time, and also allow NPCs to be more complex beasts - so long as an NPC was selected for a quest according to some criteria, e.g. personality / alignment...

Scripting differences in quests and branches and links between quest elements could produce great results if done well, but it would be a very hard design challenge. I'd enjoy this solution too, I'm sure, but it does put more of an emphasis on ongoing design - every new quest / quest section would have to be carefully constructed to fit well with others.

Personally I prefer the idea of setting up (and testing, tweaking, redesigning...) a good initial design, and letting quests emerge from there. That way (in theory), you can add content later just by expanding the options available to NPCs or factions... and everything will compensate automatically (if it is very well designed). Providing the potential for tens of hours more content could mean adding only one additional NPC action (which of course might break everything if things aren't well set up).

Whether you get engaging gameplay this way is the question - at least presuming you can implement it all. I think the resulting gameplay would be good for RPG/strategy players, but perhaps not for pure RPGers. A pre-scripted quest element combination might well work better for anti-strategists.

Neither do I, which is why I didn't...

Ok - sorry. I was inferring your meaning from the context rather than reading what you'd said carefully.
Most of what I've been saying has little to do with dynamic content generation, as you say - the content is only "generated" in the sense that combinations of needs / actions occur to "generate" quests, and combinations of pre-written dialogue (with variables) are arranged based on hand-written rules.


Crufty:
I don't think that procedurally generating the world / towns etc. is a good idea - not unless they are procedurally produced according to fairly strict parameters, then handcrafted afterwards.
It would be possible to procedurally generate the game world at the start of each game, but I think once you get into that you're really starting to leave RPGness behind, and heading for strategy / 4X (which again, I'd enjoy, but some certainly wouldn't).

I don't think many players would get the same satisfaction from exploring the world and meeting NPCs if the world and NPCs were completely procedurally generated - unless it was done more convincingly than current technology would allow.

Generating basic content before hand crafting the results is different of course. That should save a lot of time on mundane aspects, and allow designers to be creative with the more important points. This is definitely something to persue, but it does require versatile generation tools - or else designers will spend as much time fixing generated problems as they would have designing it from scratch.
This has nothing directly to do with gameplay - although it might allow more development time to be spent on gameplay issues.

 

[Mangler]

My My, much talk.

In short, all the parts we need for such a system have already been done.

Civilivation + Black&White/Populous + Nethack/Rogue/Angband + Chat A.I.'s

Then why have there been few commercial games with such a system?

Longevity,

Would you like to live a very long but boring life?

OR a very short but very exciting one?

Also whats harder to design and impliment (and cheaper)... surprise!

 

[Xi]

This entire concept seems to desire a game that never ends. Creating a world that never becomes old. The escapist's ultimate dream.

Like many others have mentioned. How can something as dynamically generic as a procedural design create memorable experiences that do not repeat or show signs of content overlap.

I believe that the future of cost effective game design lies in focused Middleware packages. In terms of AI, a few competing companies can continue to progress AI leaving Developers with the opportunity to exploit possiblities rather then waste time reinventing the wheel with every game.

In the end, procedural content relies on advanced Middleware packages. Developers will need the extra time these can generate to create extreme amounts of content and layer the content with ridiculous amounts of depth. These middleware packages also give developers a chance to develop skillsets with familiar tools. Instead of restarting each time, they have learned from mistakes and can grow in expertise and implementation. There is more time for mastery which can drive development in new ways.

The decline in gaming has come from the monumental task that video games present. When game design was limited by weaker hardware, there was more headroom to expand in the content arena. In modern times, most development studios spend 65% of there resources on technology and 35% on game design and content.(Take that with a grain of salt - like any random statistic someone pulls out of there ass)

Anyway, procedural content seems to be the next step in random generation. The difference is that there is more intelligence applied to the design. The question is how can it ever be achieved given the state of the industry right now?

 

[Section8]

This entire concept seems to desire a game that never ends. Creating a world that never becomes old. The escapist's ultimate dream.

That's about the sum of it.

Like many others have mentioned. How can something as dynamically generic as a procedural design create memorable experiences that do not repeat or show signs of content overlap.

Well, it's not as though games with pre-scripted content don't do the same. Bioware have basically been reusing the same combat model since Baldur's Gate; the D&D license gets flogged like a dead horse; Action RPGs are still trying to emulate Diablo; we see the same plotlines over and over; etc. To be perfectly honest, "memorable experiences" have been pretty few and far between in RPGs (and games in general.)

Procedural content, on the other hand, if done well, has potential to create unpredictable events and situations that don't fit the cliches, and can ultimately provide an endless series of goals to strive for, even when the here-and-now becomes dull.

I believe that the future of cost effective game design lies in focused Middleware packages. In terms of AI, a few competing companies can continue to progress AI leaving Developers with the opportunity to exploit possiblities rather then waste time reinventing the wheel with every game.

Not necessarily. A company like Bethesda that specialises in producing incremental advancement of similar games that all have the same basic technical requirements. It's really in their best interest to come up with general solutions that can be re-used and built upon with each new game.

That's why EA have their own internal teams developing tools and features that can be applied across the whole range of their EA sports titles. It's very similar to middleware, but it's taking the initiativ and recognising a need rather than waiting for someone else to develop the tech and then licensing it for a fee.

 

[Human Shield]

I think emergent properties can come from higher level abstractions.

You don't need AI for each NPC. You can keep track of a variables like "crime rate" which will effect the chance of the player encounting a theif or crime, the number can change if you start killing guards or criminals.

It is a encounter roll that can use templates to create some crime happening. It doesn't eat up CPU looking at every NPC, it could spawn generic NPCs out of visual range. Most importantly the number can be effected by the player and other variables like population numbers; or link with other towns and empires including population and leadership desire to crack down on crime.

It is number crunching like a 4X game that AIs can do but it is playing around with itself, creating an enviroment that the player can alter.

 

[Xi]

Well, it's not as though games with pre-scripted content don't do the same. Bioware have basically been reusing the same combat model since Baldur's Gate; the D&D license gets flogged like a dead horse; Action RPGs are still trying to emulate Diablo; we see the same plotlines over and over; etc. To be perfectly honest, "memorable experiences" have been pretty few and far between in RPGs (and games in general.)

Procedural content, on the other hand, if done well, has potential to create unpredictable events and situations that don't fit the cliches, and can ultimately provide an endless series of goals to strive for, even when the here-and-now becomes dull.

Good point. I guess I find it hard to believe that developers will be able to create enough depth with such a randomly and sometimes chaotic system. Though the sheer number of shallow, yet unique, instances may be enough to please most of us. Quantity over quality debate to a degree.

Also, gameplay has been on the decline ever sense graphical splendor took over. Eye Candy drives many industries(porn, though there can be mild participation like in modern video games, being a good example.). It's kind of a mixed bag because I can see the merits of good graphics, I just hate to see graphics get more development time then gameplay. Hell, graphics alone do nothing(unless theres nudity!) but a text based game with tons of gameplay content can still be an awesome experience.

Good graphics are merely a form of artificial masturbation. It's superficial in comparison to the actual game. All it can do is promote the feeling of experience, but it isn't actually the experience. Gameplay is the real reason to play a game in the first place. You experience the game through its gameplay.... Ok, I've gone off on a tanget...

Not necessarily. A company like Bethesda that specialises in producing incremental advancement of similar games that all have the same basic technical requirements. It's really in their best interest to come up with general solutions that can be re-used and built upon with each new game.

That's why EA have their own internal teams developing tools and features that can be applied across the whole range of their EA sports titles. It's very similar to middleware, but it's taking the initiativ and recognising a need rather than waiting for someone else to develop the tech and then licensing it for a fee.

The only problem I see with this is that it promotes 'Sequel Syndrome.' I'm not sure whether that's a good or bad thing, but I'm sure a lot of people would argue against it. It's a good example of how a company can expand on there design capability though. Good point.